Dataset Viewer
Auto-converted to Parquet Duplicate
sample_id
stringclasses
10 values
image
imagewidth (px)
1.28k
1.28k
instruction
stringclasses
10 values
website
stringclasses
10 values
gold_bbox
listlengths
4
4
action_json
stringclasses
5 values
previous_actions
listlengths
1
19
task_id
stringclasses
10 values
0714_X_2_1#2
λŸ¬λ‹μ„ μ‹œμž‘ν•˜λ €κ³  ν•˜λŠ”λ° λ‹€λ₯Έ μ‚¬λžŒλ“€μ΄ μ–΄λ–€ λŸ¬λ‹ν™”λ₯Ό μ‹ λŠ”μ§€ λ“£κ³  μ‹Άμ–΄. Xμ—μ„œ λŸ¬λ‹ν•˜λŠ” λΆ„λ“€μ—κ²Œ μ–΄λ–€ λŸ¬λ‹ν™”λ₯Ό μ‹ λŠ”μ§€ λ¬»λŠ” 글을 올렀쀘.
X_2
[ 51.5, 619.5, 601.4, 706.8 ]
{"operation": "CLICK", "value": null, "target_text": "이메일 λ˜λŠ” μ‚¬μš©μž 이름"}
[ "GOTO https://x.com/" ]
0714_X_2_1
0720_zigzag_1#20
여름에 νŽΈν•˜κ²Œ μž…μ„ 검은색 λ°˜νŒ” μ›ν”ΌμŠ€λ₯Ό μ°Ύμ•„μ£Όμ„Έμš”. μ§€κ·Έμž¬κ·Έμ—μ„œ λ°˜νŒ” μ›ν”ΌμŠ€λ₯Ό κ²€μƒ‰ν•˜κ³  색상은 λΈ”λž™, 가격은 5만 원 μ΄ν•˜λ‘œ μ„€μ •ν•΄μ£Όμ„Έμš”. ꡬ맀 ν›„κΈ°κ°€ μΆ©λΆ„ν•˜κ³  λ¬΄λ£Œλ°°μ†‘μ΄ κ°€λŠ₯ν•œ μƒν’ˆ ν•˜λ‚˜λ₯Ό 골라 μ†Œμž¬μ™€ 총기μž₯, 세탁 방법을 ν™•μΈν•œ λ’€ μ°œν•΄μ£Όμ„Έμš”
zigzag
[ 297.8, 191.4, 519, 600.7 ]
{"operation": "CLICK", "value": null, "target_text": null}
[ "GOTO https://zigzag.kr/", "CLICK", "CLICK 카카였둜 3초 λ§Œμ— μ‹œμž‘ν•˜κΈ°", "CLICK", "TYPE \"[ID]\" into input", "CLICK λΉ„λ°€λ²ˆν˜Έ μž…λ ₯", "CLICK", "TYPE \"[PASSWORD]\" into input", "HOTKEY ['Enter']", "CLICK ν™ˆ", "CLICK", "CLICK", "TYPE \"λ°˜νŒ” μ›ν”ΌμŠ€\" into input", "HOTKEY ['Enter']", "SCROLL down", "SCROLL up", ...
0720_zigzag_1
0714_ebsi_1#2
λ‚˜μ—κ²Œ λ§žλŠ” 인터넷 κ°•μ˜ μ„ μƒλ‹˜μ„ μ°Ύκ³  μ‹Άμ–΄. ebsiμ—μ„œ 고등학ꡐ 3ν•™λ…„ 미적뢄 κ°•μ’Œλ₯Ό 검색 ν›„ κ°€μž₯ μœ„μ— λ‚˜μ˜€λŠ” κ°•μ’Œλ₯Ό μ°œν•΄μ€˜.
ebsi
[ 767.9, 644.2, 824.7, 665.3 ]
{"operation": "CLICK", "value": null, "target_text": "λ‹«κΈ°"}
[ "GOTO https://www.ebsi.co.kr/ebs/pot/poti/main.ebs" ]
0714_ebsi_1
0714_google_docs_1#2
비타민C의 효λŠ₯에 κ΄€ν•œ 글을 적고 μ‹Άμ–΄. ꡬ글 λ…μŠ€μ—μ„œ 비타민C의 효λŠ₯에 κ΄€ν•œ 글을 적어주고 pdf둜 λ‹€μš΄λ‘œλ“œ ν•΄μ€˜.
google_docs
[ 677.2, 151.7, 1231.4, 226.2 ]
{"operation": "CLICK", "value": null, "target_text": "이메일 λ˜λŠ” νœ΄λŒ€μ „ν™”"}
[ "GOTO https://accounts.google.com/v3/signin/identifier?continue=https%3A%2F%2Fdocs.google.com%2F&dsh=S-1306464316%3A1784019949610766&emr=1&followup=https%3A%2F%2Fdocs.google.com%2F&osid=1&passive=1209600&flowName=GlifWebSignIn&flowEntry=ServiceLogin&ifkv=Ac50bxu4zUBnseXWeGwhxf6scCIFMgepP6dAgTttSf-N5tUkLqp3QMfTLehfz...
0714_google_docs_1
0714_instagram_2_1#2
μΈμŠ€νƒ€κ·Έλž¨μ—μ„œ 'kimsseungmini' 계정에 였늘 λ§Œλ‚˜μ„œ 정말 λ°˜κ°€μ› λ‹€λŠ” λ‚΄μš©μ˜ λ©”μ„Έμ§€λ₯Ό 2μ€„λ‘œ λ³΄λ‚΄μ€˜.
instagram_2
[ 259.1, 347.8, 998.9, 402.3 ]
{"operation": "CLICK", "value": null, "target_text": null}
[ "GOTO https://www.instagram.com/" ]
0714_instagram_2_1
0714_nate_1#2
μ—¬ν–‰κ³„νšμ„ μ§°λŠ”λ° 메일에 μ €μž₯해두고 μ‹Άμ–΄. λ„€μ΄νŠΈμ—μ„œ λ‚΄κ²Œμ“°κΈ°λ‘œ μ—¬ν–‰κ³„νš 메일을 μž‘μ„± ν›„ μ „μ†‘ν•΄μ€˜.제λͺ©μ€ μ—¬ν–‰κ³„νšμœΌλ‘œ
nate
[ 873.9, 283, 1109, 310.9 ]
{"operation": "CLICK", "value": null, "target_text": null}
[ "GOTO https://www.nate.com/" ]
0714_nate_1
0714_naver_webtoon_1#2
넀이버 μ›Ήνˆ°μ—μ„œ λ¬΄ν˜‘ 검색 ν›„ μƒμœ„ 5개 μž‘ν’ˆ 관심등둝 ν•˜κΈ°
naver_webtoon
[ 1063.4, 17.5, 1101.9, 34.2 ]
{"operation": "CLICK", "value": null, "target_text": "둜그인"}
[ "GOTO https://comic.naver.com/index" ]
0714_naver_webtoon_1
0714_ohou_se_1#3
κ±°μ‹€μ—μ„œ μ‚¬μš©ν•  러그λ₯Ό 사렀고 ν•˜λŠ”λ° μ–΄λ–€κ±Έ 골라야할지 λͺ¨λ₯΄κ² μ–΄. 였늘의 μ§‘μ—μ„œ 러그λ₯Ό 검색 ν•΄μ£Όκ³  λ°©μˆ˜κ°€ κ°€λŠ₯ν•œ μ œν’ˆμœΌλ‘œ 3개 μ°œν•΄μ€˜.
ohou_se
[ 694, 580, 752.1, 634 ]
{"operation": "CLICK", "value": null, "target_text": null}
[ "GOTO https://ohou.se/", "CLICK 둜그인" ]
0714_ohou_se_1
0714_pexels_1#2
pexelsμ—μ„œ λ°”λ‹€ 검색 ν›„ λΉ„κ΅ν•˜κ³  λ‹€μš΄λ‘œλ“œ ν•˜κΈ°
pexels
[ 144.5, 265.7, 631.7, 305.8 ]
{"operation": "CLICK", "value": null, "target_text": null}
[ "GOTO https://www.pexels.com/ko-kr/" ]
0714_pexels_1
0714_pinterest_1#2
κ·€μ—¬μš΄ 고양이 사진을 λ‹€μš΄λ‘œλ“œ ν•˜κ³ μ‹Άμ–΄. ν•€ν„°λ ˆμŠ€νŠΈμ—μ„œ 고양이λ₯Ό κ²€μƒ‰ν•˜κ³  κ°€μž₯ 첫 λ²ˆμ§Έμ— λ‚˜μ˜€λŠ” 사진을 λ‹€μš΄λ‘œλ“œ ν•΄μ€˜.
pinterest
[ 984.5, 22.9, 1093.6, 91.7 ]
{"operation": "CLICK", "value": null, "target_text": "둜그인"}
[ "GOTO https://kr.pinterest.com/" ]
0714_pinterest_1

πŸ‡°πŸ‡· Ko-MultimodalMind2Web

An offline Korean web-agent benchmark, built manually by native Korean annotators.

πŸ“‹ 549 tasks Β· πŸ–ΌοΈ 12,905 scored steps Β· 🌐 144 websites Β· ✍️ ~480 annotator-hours Β· 🧭 Offline evaluation

πŸ“Œ Overview

Ko-MultimodalMind2Web is an offline web-agent benchmark for Korean websites. It contains 549 tasks and 12,905 scored action steps across 144 websites, collected from human browser demonstrations. The published task interface follows Multimodal Mind2Web: given a Korean task, the current screenshot, and preceding action history, an agent predicts the next operation, target, and action value. Evaluation reports element accuracy, step success, and task success.

✍️ Built by Human Annotators

All tasks and reference demonstrations were created manually by Korean human annotators. They designed realistic requests, operated the websites themselves, and recorded browser demonstrations of each task.

Item Details
πŸ‘€ Annotators 2 native Korean speakers
⏱️ Total effort About 480 annotator-hours

This is the shared collection and annotation effort behind all three Korean releases in the suite below. The estimate is based on two annotators working four hours per day, five days per week, for approximately three months (about twelve weeks).

The benchmark follows the offline task/trajectory/action organization used by Mind2Web and Multimodal Mind2Web / SeeAct: language instructions and screenshots are paired with next-action targets. The Korean task demonstrations and annotations were collected manually. The released operation inventory, action history, and accepted alternatives are explicitly retained so evaluation is reproducible on this collection.

How the benchmark was constructed and verified

  1. Human task design and demonstration. Korean annotators created realistic tasks on Korean websites and Korean-language interfaces, then performed the complete tasks themselves. Browser events, screenshots, and target-element information were captured from their actual interactions.
  2. Step-level annotation. The recorded demonstrations were organized into task-linked steps with an operation, argument/value, target geometry when relevant, and the preceding action history. Both original and concise task-instruction variants are supplied.
  3. Manual target correction. Native browser dialogs, separate popup windows, operating-system select menus, and dynamically replaced controls were inspected using their screenshots. Missing or incorrectly aligned targets were corrected manually. Selection actions were checked against subsequent screenshots to distinguish the selected value from a transient highlight.
  4. Alternative-answer review. Equivalent form-submission actions were represented as alternative valid answers. The final release includes 350 manually located alternative CLICK targets, together with template-matched alternatives and valid Enter-key submissions. Operation, value, and location must belong to the same accepted answer candidate.
  5. Consistency and scoring checks. All 12,905 scored-step IDs are unique; all steps resolve to one of the 549 tasks and agree with task-level step counts. All 7,197 primary target boxes are finite, have positive area, and lie within their screenshot dimensions. Gold-label round-trip checks return 1.0 for element accuracy, step success, and task success using the included scorer.

πŸ“Š Dataset statistics

Recorded operation distribution

Category Count Share
CLICK 5,879 45.56%
TYPE 1,399 10.84%
HOTKEY 444 3.44%
SCROLL 5,007 38.80%
CLEAR 15 0.12%
GO_BACK 134 1.04%
DRAG 27 0.21%

The 549 human reference trajectories contain 14,042 recorded steps. Offline next-action scoring is restricted to the seven interaction operations listed above. URL navigation (GOTO), waiting (WAIT), and task termination (FINISH) are excluded from step-level scoring, leaving 12,905 scored interaction steps. statistics.json includes per-website counts, task categories, trajectory-length statistics, and alternative-action counts.

🧩 KORMo-VL Benchmark Suite

This dataset is part of the KORMo-VL Korean GUI-agent benchmark suite. Each release is built from manually collected Korean data and follows the protocol of an established international benchmark.

Dataset Setting Follows
Ko-Screenspot Single-image GUI grounding ScreenSpot
Ko-MultimodalMind2Web Offline next-action prediction Multimodal Mind2Web
Ko-Online-Mind2Web Live task completion Online Mind2Web

Download and preview

This repository is private. Authenticate with a Hugging Face account that has access to the repository.

from huggingface_hub import snapshot_download

snapshot_download(
    "KORMo-VL/Ko-MultimodalMind2Web",
    repo_type="dataset",
    local_dir="Ko-MultimodalMind2Web",
    allow_patterns=["README.md", "statistics.json", "evaluate.py", "multimodal_mind2web_ko.tar.gz"],
    token=True,
)

The dataset viewer contains 10 illustrative examples. The complete benchmark is provided in the separate full-data file listed above. These examples are selected from the full data.

from datasets import load_dataset

preview = load_dataset("KORMo-VL/Ko-MultimodalMind2Web", split="test", token=True)

Extract the full data from the downloaded repository directory:

cd Ko-MultimodalMind2Web
tar -xzf multimodal_mind2web_ko.tar.gz

Evaluation

Use the included evaluate.py to report element accuracy, step success rate, and task success rate on the released scored steps. The reference scorer supports accepted alternative actions.

Contact

  • KyungTae Lim, Professor at KAIST. ktlim@kaist.ac.kr

Contributor

Citation

References for KORMo and the benchmark protocol followed by this release:

@misc{KORMo,
  author = {Minjun Kim and Hyeonseok Lim and Hangyeol Yoo and Inho Won and Seungwoo Song and Minkyung Cho and Junghun Yuk and Changsu Choi and Dongjae Shin and Huije Lee and Hoyun Song and Alice Oh and KyungTae Lim},
  title = {KORMo: Korean Open Reasoning Model for Everyone},
  year = {2025},
  publisher = {GitHub},
  journal = {Technical Report},
  url = {https://arxiv.org/abs/2510.09426}
}
@inproceedings{zheng2024seeact,
  title = {GPT-4V(ision) is a Generalist Web Agent, if Grounded},
  author = {Boyuan Zheng and Boyu Gou and Jihyung Kil and Huan Sun and Yu Su},
  booktitle = {Forty-first International Conference on Machine Learning},
  year = {2024},
  url = {https://openreview.net/forum?id=piecKJ2DlB}
}
@inproceedings{deng2023mind2web,
  title = {Mind2Web: Towards a Generalist Agent for the Web},
  author = {Xiang Deng and Yu Gu and Boyuan Zheng and Shijie Chen and Sam Stevens and Boshi Wang and Huan Sun and Yu Su},
  booktitle = {Advances in Neural Information Processing Systems},
  volume = {36},
  pages = {28091--28114},
  year = {2023},
  publisher = {Curran Associates, Inc.},
  url = {https://proceedings.neurips.cc/paper_files/paper/2023/file/5950bf290a1570ea401bf98882128160-Paper-Datasets_and_Benchmarks.pdf}
}
Downloads last month
-

Collection including KORMo-VL/Ko-MultimodalMind2Web

Papers for KORMo-VL/Ko-MultimodalMind2Web